-
ODDITY LEARNING IN HONEYBEES (APIS MELLIFERA)
A THESIS SUBMITTED TO THE GRADUATE DIVISION OF THE UNIVERSITY OF
HAWAI‘I AT MĀNOA IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF
MASTER OF ARTS
IN
PSYCHOLOGY
DECEMBER 2012
By
Nicole M. Muszynski
Thesis Committee:
Patricia Couvillon, Chairperson
Leonidas Doumas
Scott Sinnett
Keywords: learning, honeybees, concepts, invertebrates,
oddity
-
ii
In dedication to Shane Wilson-South.
-
iii
Acknowledgements
I owe a great deal of gratitude to those who helped me write,
discuss, and
complete my thesis. Without the help of my advisor, Patricia
Couvillon, this
thesis would still be in its infancy. I owe many thanks to her
for allowing me the
chance to work with her and for inspiring and motivating me to
accomplish my
goals. I also would like to give another special thanks to
Gentaro Shishimi for his
input and insights. In addition, I would like to thank all the
undergraduates,
Brandon Balangue, Charmaine Alontaga, and Karl Alcover, for
their hard work
and for making the laboratory a very lively place. I would also
like to thank my
committee members, Leonidas Doumas and Scott Sinnett, for their
time, input,
and encouragement.
-
iv
Table of Contents
Acknowledgements………………………………………………………….………..…iii
List of Tables………………………………………………………………...…..…..…..vi
List of Figures………………………………………………………………...….....…..vii
Introduction…………………………………………………………….……....…….….8
Historical Overview………………………………………………………………8
Invertebrate Learning…………...……………………………………………….10
Associative Versus Cognitive Explanations……………….……………………..12
Honeybee Concept Learning………………………………………………..……15
Oddity Learning…………………………………………………………….……17
Experiment
1……………………………………............................................................21
Introduction………………………………………………………………………21
Subjects..………..……………………………………………………..…………21
Apparatus & Stimuli……………………………………………………..………22
Procedure…………………………………………………………………….….23
Pretraining...……………….....................................................................23
Training………………...……………………………………….….……23
Results & Discussion………………………………………………………….…24
Experiment 2………..………………………………..…………………………………29
Introduction………………………………………………………………………29
Subjects……………………………………………..……………………………29
Apparatus & Stimuli………………………………………….………….………29
-
v
Procedure……………………………………………………………….….…….30
Pretraining………...…………………………………………..…………30
Training………………………………...…………………...……...……30
Results & Discussion………………………………………………….…………30
Experiment 3………..………………………………..…………………………………33
Introduction………………………………………………………………………33
Subjects……………………………………………..……………………………33
Apparatus & Stimuli………………………………………….………….………34
Procedure……………………………………………………………….….…….35
Pretraining………...…………………………………………..…………35
Training………………………………...…………………...……...……35
Results & Discussion………………………………………………….…………36
Experiment 4………..………………………………..…………………………………38
Introduction………………………………………………………………………38
Subjects……………………………………………..……………………………38
Apparatus & Stimuli………………………………………….………….………38
Procedure……………………………………………………………….….…….39
Pretraining………...…………………………………………..…………39
Training………………………………...…………………...……...……39
Results & Discussion………………………………………………….…………39
General Discussion………..…………………...……..…………………………………42
References………………………………..………………………………………...……47
-
vi
List of Tables
Table 1…………………………………………………………………………………...46
-
vii
List of Figures
Figure 1……………………………………………………………………...……….…..22
Figure 2……………………………………………………………………..……………25
Figure 3……………………………………………………………………………….….31
Figure 4……………………………………………………………………………….….34
Figure 5……………………………………………………………………………….….36
Figure 6……………………………………………………………………………….….40
-
8
Introduction
“I know, indeed, that brutes do many things better than we do,
but
I am not surprised at it; for that, also, goes to prove that
they act
by force of nature and by springs, like a clock, which tells
better
what the hour is than our judgment can inform us” (Rene
Descartes, 1646).
Historical Overview
Descartes attributed to humans both voluntary action and
involuntary reflexes while to nonhuman animals only innate
involuntary reflexes (1637). This idea persisted
until Charles Darwin published his ideas on evolution in the two
books, On the Origin of
Species (1859) and The Descent of Man (1871), where he
speculated about the mental
continuity between humans and nonhuman animals. To provide
evidence of mental
continuity, George Romanes, Darwin’s pupil and collaborator,
published Animal
Intelligence in 1892. Even though Romanes relied heavily on
anecdotes and
anthropomorphic interpretations, he nonetheless attempted to
characterize the evolution
of intelligent behavior. He provided observational accounts of
both invertebrates (e.g.,
scorpions, bees, ants, mollusks, including oysters and
octopuses) and vertebrates (e.g.,
fishes, reptiles, and mammals) with particular emphasis on cats,
dogs, and elephants
(Romanes, 1892).
Conwy Lloyd Morgan, Romanes’ successor, was more critical than
Romanes and
rejected anecdotes and anthropomorphic interpretations in favor
of empirical evidence.
He conducted field experiments on reflexes with several species,
including young chicks,
ducklings, and dogs. Morgan is best known, however, for his
cautionary statement, “In no
case may we interpret an action as the outcome of the exercise
of a higher psychical
faculty if it can be interpreted as the outcome of the exercise
of one which stands lower in
-
9
the psychological scale” (Morgan, 1894). This statement became
known as Morgan’s
Canon, a restatement of Occam’s Razor in psychological
terms.
Also, in response to Romanes’ anthropomorphism and lack of
experimentation,
Edward Thorndike brought the study of animal intelligence into
the laboratory.
Thorndike viewed the anecdotal observations of animal
intelligence as “eulogies,” that is,
an emphasis on the unusual and the marvelous animal behavior
rather than on the normal
and the unremarkable (Thorndike, 1898). To remedy this situation
he built “puzzle
boxes” from which experimental subjects, such as cats, dogs, and
young chicks, had to
escape. He then measured escape latencies in a series of trials
with each species of
subject, all of which exhibited the same results. The escape
latencies gradually decreased
over trials. The subjects never exhibited spontaneous or
insightful learning as would be
indicated by an immediate and sharp decrease in escape time.
From these experiments,
Thorndike established the Laws of Effect and Exercise, two
cornerstones of associative
learning (1911). The rigorous experimental methods Thorndike
used to study animal
learning became the foundation for the study of instrumental
conditioning, later called
operant conditioning by Skinner (1938), a procedure where
responses are followed by
reward or punishment.
At the same time Thorndike was conducting his experiments, Ivan
Pavlov was
conducting experiments on the digestive reflexes of dogs
(Pavlov, 1927). Accidentally,
he discovered the conditioned reflex. When he noticed his dogs
salivated before any food
had been placed in their mouths, he concluded that a novel
stimulus paired with an
unconditioned reflex elicited a conditioned reflex. In his
subsequent work, he attempted
to characterize the parameters necessary to produce conditioned
reflexes. Pavlov’s work
later became the foundation for classical or Pavlovian
conditioning. Both Pavlovian and
instrumental conditioning procedures provide powerful tools for
the laboratory study of
learning.
-
10
Invertebrate Learning
Initially, research was conducted on a wide variety of species
to determine the range of species that could learn. However, once
it was established that all species likely
could learn, subsequent research on learning focused primarily
on the rat and pigeon
(Beach, 1950; Bitterman, 2006). Both the rat and the pigeon were
considered ideal
candidates for research subjects in studies of learning because
they are abundant, cheap,
hardy, and easily reared for the laboratory (Beach, 1950;
Bitterman, 2006). The work
with these species served to facilitate the development of
general learning principles,
assumed to apply to all species. Later, technological advances
in cellular neuroscience
facilitated a renewed interest in the learning of other species,
including invertebrates.
Invertebrates such as gastropods (e.g., aplysia) and crustaceans
(e.g., crayfish) proved
ideal for studies of the biological basis for learning because
they have relatively simple
nervous systems with very large and fast neurons (Corning, Dyal,
& Willows, 1973).
Interestingly, while gastropods and crustaceans make ideal
subjects for
electrophysiological experiments, they are not ideal subjects
for learning experiments
because of their limited sensory and motor repertoires. For this
reason, invertebrate
learning studies have focused primarily on cephalopods (e.g.,
octopuses) and insects
(e.g., honeybees). These species are not without limitations for
learning work, although,
arguably, the octopus presents more methodological problems than
the honeybee (see:
Corning, Dyal, & Willows, 1975; Bitterman, 1975; Boal,
1996). Octopuses are not hardy
in the lab and require special housing and proper handling.
Conversely, honeybees are
easier to maintain for laboratory study. Furthermore, honeybees
have a range of sensory,
motor, and motivational capabilities that make them ideal
subjects for learning research.
Karl von Frisch (1950) conducted experiments to determine the
sensory
capacities of the honeybee. He found that honeybees have color
vision similar to that of
humans, although they do not detect longer wavelengths (e.g.,
red) and do detect shorter
wavelengths (e.g., ultraviolet). Honeybees have chemoreceptors
on their antennae and
tarsi (von Frisch, 1950; Scheiner, Page, & Erber, 2001)
which are important for kin
-
11
discrimination, recognition of floral odors, and perception of
pheromones (Robertson &
Wanner, 2006). Honeybees also detect magnetic fields (Gould,
Kirschvink, & Deffeyes,
1978; Walker & Bitterman, 1985) as well as vibration and
touch (Kirchner, 1993; Tautz
& Rohrseitz, 1998; Rohrseitz & Tautz, 1999; Nieh &
Tautz, 2000). The motor
capabilities of honeybees include flying, walking, oscillating
antennae, as well as stinger
and proboscis extension, all of which provide a number of
response systems for learning
experiments. In addition, motivated for nectar (sugar), foraging
honeybees will
consistently return to a high concentration of sugar solution
and will do so without
satiating since they are able to regurgitate their food at the
hive in order to make room for
more.
In early experiments on learning, researchers were mainly
interested in exploring
basic learning phenomena in the honeybee using analogs of
vertebrate learning
experiments. Two procedures were developed for studying learning
in honeybees, the
proboscis extension and the free-flying techniques. The
proboscis extension technique
uses foraging honeybees that are caught and then restrained,
allowing only minimal head
movement. When a bee’s antennae are touched with sucrose, the
proboscis extends, a
response referred to as an unconditioned reflex. This
unconditioned reflex can be paired
with a novel stimulus such as odor to elicit a conditioned
reflex of proboscis extension.
Traditionally, proboscis extension has been used to explore
basic Pavlovian conditioning
and has provided much evidence that the conditioning of
honeybees is similar to
vertebrates (Bitterman, Menzel, Fietz, & Schäfer, 1983;
Batson, Hoban, & Bitterman,
1992; Couvillon, Hsiung, Cooke, & Bitterman, 2005). The
proboscis extension technique
has been useful for the study of Pavlovian conditioning, but it
does have limitations. The
experimental bees must be restrained which permits satiation,
and only odors or touch
applied to the antennae can serve as the conditioned stimuli. To
date, visual stimuli are
not effective as conditioned stimuli for a conditioned response
of proboscis extension.
Due to these limitations, the free-flying technique is generally
regarded as more powerful
for conducting learning experiments with honeybees.
-
12
The free-flying technique uses an individual foraging honeybee
trained to visit a
laboratory window or table to feed on artificial flowers. The
bee must learn about the
flower’s color or scent in order to receive a high concentration
of sucrose reward. After it
feeds, the bee is free to return to the hive and then again to
the laboratory. This method
has been used to study acquisition, extinction, parameters of
reward, and choice
discrimination in honeybees (Couvillon & Bitterman, 1985;
Couvillon & Bitterman,
1988; Couvillon, Lee, & Bitterman, 1991; Bitterman, 1996),
and, as with the proboscis
extension technique, the results are similar to those of
vertebrates.
Such similarities in general learning phenomena are remarkable
given that
vertebrates and invertebrates are believed to have shared a
common ancestor some half a
billion years ago. Furthermore, the brain structures of
honeybees are very different than
those of vertebrates. The similarities might make sense,
however, if the biological
mechanisms of learning occur at the cellular level (Kandel &
Hawkins, 1992). Albeit,
convergent evolution is perhaps more likely, that is, “different
phenomena may be
produced by the same processes, and what appear to be identical
phenomena may be
produced by different processes” (Bitterman, 1975).
Associative Versus Cognitive Explanations
In the spirit of Thorndike and Pavlov, the work with honeybees
has concentrated
on associative learning principles in order to facilitate
comparison with vertebrates.
Thorndike and Pavlov’s associative approach, in fact,
characterized the development of
the field of animal learning with the exception of some early
work with primates on
insight, problem-solving, and language. However, there was work
with other vertebrate
animals that hinted at cognitive capacities beyond associative
learning.
Although later discredited, Wolfgang Kohler (1918) conducted
experiments on
chickens and chimpanzees and proposed a cognitive rather than an
associative
explanation of discrimination learning. In his experiments,
subjects were trained with
reward to choose a light gray stimulus (S+) instead of a darker
gray stimulus (S-) drawn
-
13
from stimuli on a black to white continuum. After training, a
transposition test was then
presented to the subjects in which they had to choose between
the original gray stimulus
(S+) and a new, even lighter, stimulus along the black to white
continuum. He observed
that the subjects chose the lighter stimulus twice as often,
leading him to conclude that
the chickens were capable of learning “togetherness” or
“relations” (Kohler, 1918).
Later, in direct opposition to Kohler’s (1918) relational
account of discrimination
learning, Kenneth Spence (1936) proposed the first formal theory
of discrimination
learning. He claimed that Kohler’s experimental subjects would
consistently choose the
lighter stimulus, not based on the ability to learn about
relations, but on the absolute
properties of the stimuli. Spence believed animals acquired an
excitatory (E) tendency to
choose a rewarded stimulus and an inhibitory (I) tendency to
avoid a nonrewarded
stimulus; both tendencies generalize to other stimuli on the
same stimulus continuum and
performance is determined by E - I. Spence (1936) concluded
that, at least in theory,
relational learning was not necessary to account for Kohler’s
results.
Spence’s theory of discrimination learning provided further
support for
associative explanations of animal learning. Not until Edward
Tolman, a major figure in
the “cognitive revolution,” published a paper in 1948 theorizing
about the spatial
mapping abilities of rats was there a renewed interest in
cognitive explanations of
learning. In his seminal paper, he conducted a series of
experiments on maze learning in
rats and concluded that they had learned to correctly navigate,
not on the basis of
associative learning responses, but rather on the basis of
“cognitive maps.” Tolman’s idea
of “cognitive maps” was later shown to be an inadequate
explanation of the rats’ maze
learning abilities. Instead of relying on “cognitive maps” the
rats were relying on extra-
maze cues such as visible landmarks that could be seen from
inside the maze and could
serve to guide navigation. Regardless, his ideas signaled a
change in perspective from an
associative explanation to a cognitive explanation of
learning.
Tolman’s work on spatial learning in rats inspired a lot of
research on memory
using mazes, particularly the radial arm maze (Olton &
Samuelson, 1976). Traditionally
developed for the use of rats, a typical radial arm maze has 8
protruding arms from a
-
14
center circle. In studies of spatial learning, all the arms are
baited in a trial, and, in order
to receive food reward, the rat must visit all of the arms. An
error in memory is a revisit
to an empty arm. Rats were able to successfully navigate the
maze with limited revisits,
suggesting they have spatial short-term memory.
Just as Tolman’s ideas inspired work using the radial arm maze
to explore rats’
spatial memory, his ideas also inspired work using the
matching-to-sample (MTS)
procedure to explore pigeons’ short-term memory (Blough, 1959).
In a typical MTS
procedure, a sample stimulus is presented for a predetermined
time, then there is a
simultaneous presentation (or sometimes a successive
presentation) of two choice stimuli,
one different from and one the same as the sample. In order for
a subject to receive a food
reward, a correct choice must be made to the choice-stimulus
that is the same as the
sample stimulus. This procedure and its variation,
nonmatching-to-sample (NMTS), have
also proven useful for studying concept learning. (Note that the
term concept is used
throughout this thesis to indicate learning that may not be
explained with basic principles
of association.)
Along with the MTS and NMTS procedures, concept learning is
explored using a
same/different (SD) procedure (Wright & Katz, 2006).
Typically in a SD procedure, pairs
of stimuli are judged as same or different by a subject’s choice
of corresponding “same”
or “different” response areas. In SD, MTS, and NMTS procedures,
concept learning is
assessed by transfer tests to novel stimuli, with good 1st trial
performance indicating
positive transfer (Thomas & Noble 1988). Recently, using SD,
MTS, and NMTS
procedures, evidence of concept learning has emerged in baboons,
rhesus and capuchin
monkeys, parrots, pigeons, and possibly even in honeybees
(Wright & Katz, 2006;
Chittka & Jenson, 2011). Concept learning seems to be a
general vertebrate phenomenon,
warranting an exploration of whether concept learning is unique
to vertebrates, or
whether invertebrates, like the honeybee, are capable of
learning concepts as well.
-
15
Honeybee Concept Learning
To date, there have only been four published studies exploring
whether honeybees
are capable of learning concepts. Honeybees have been trained in
discrimination
problems that include symmetry/asymmetry (Giurfa, Eichmann,
& Menzel, 1996),
same/different (Giurfa, Zhang, Jenett, Menzel, & Srinivasan,
2001), above/ below
(Avarguès-Weber, Dyer, & Giurfa, 2011), and a combination of
above/below, left/right
and difference (Avarguès-Weber, Dyer, Combe, & Giurfa,
2012). Each yielded
significant results using methodological paradigms similar to
those used with vertebrates.
Giurfa et al. (1996) conducted an experiment to determine
whether honeybees
could discriminate symmetry. Honeybees were presented with a
succession of eight
triads; each contained three different stimuli presented
simultaneously. Depending on an
individual bee’s assignment, the correct choice was the
symmetrical stimulus rather than
the different asymmetrical stimuli (A+ B- C-), or the
asymmetrical stimulus rather than
the different symmetrical stimuli. Unrewarded transfer trials
were interspersed among
training trials. The bees’ performance on the transfer trials
was similar to that on the
training trials, leading the authors to conclude that the bees
had learned to abstract the cue
“symmetry.” Whether the honeybees actually learned “symmetry,”
however, is debatable
given that transfer of learning is only indicative of concept
learning if the 1st trial
performance is above chance (Thomas & Noble, 1988) and first
trial data were not
provided.
To validate and further explore concept learning in honeybees,
Avarguès-Weber
et al. (2011) conducted an experiment using a MTS procedure to
assess whether
honeybees could master a conceptual spatial relation such as
above and below. After
training trials, the bees were presented with an unrewarded
transfer test to determine
whether concept learning had taken place. The results showed
positive transfer, however,
it could be argued that the bees did not learn above/below
concepts, but merely solved a
simple spatial discrimination problem.
-
16
Giurfa et al. (2001) performed another experiment using MTS and
NMTS
procedures to assess “same” and “different” concepts. Results of
the honeybees’
performance in both training and transfer tests were
significantly better than chance,
leading the authors to suggest the bees were learning concepts.
Although, given the
relatively large number of training trials (60 trials) that each
bee received, it is possible
the honeybees had just memorized all possible responses (only
four responses), and thus
learned a purely associative stimulus-response strategy. In
addition, 1st trial transfer
performance was not measured, leaving it unclear whether the
honeybees had learned
concepts or if generalization to new stimuli took place during
the multiple transfer trials.
Additional experiments of simultaneous above-below, left/right,
and difference
concept learning in honeybees by Avarguès-Weber et al. (2012)
yielded results similar to
the studies mentioned above. Thus, Avarguès-Weber et al (2012)
suggest, based on the
honeybees’ significantly better than chance performance and
successful transfer, that
honeybees are able to abstract concepts. This conclusion came
only after the authors
claimed to have excluded alternative explanations like template
matching, perceptual
generalizations, difference in centers of gravity, broad
orientation cues, and pixel-by-
pixel comparison. Upon closer inspection, however, the transfer
tests did not meet the
criterion to demonstrate transfer of learning as established by
Thomas and Noble (1988).
The results of all four studies, taken together, “hint” at the
possibility that
honeybees may learn “concepts.” These studies, although flawed,
also suggest that
concept learning is not unique to vertebrates. Of particular
interest here is oddity
learning, generally regarded as one of the ways to study concept
learning in nonhuman
animals (Robinson, 1933; Roitberg & Franz, 2004; Hille,
Dehnhardt, & Mauck, 2006).
-
17
Oddity Learning
Robinson (1933) is one of the first researchers to develop and
present an animal
subject (e.g., macaque monkey) with an oddity problem. In order
to test whether the
animal was capable of responding based on the relationship
between stimuli, he designed
an oddity experiment in which one odd and two identical nonodd
stimuli were
simultaneously presented to a subject. To ensure that the
successful solution of the
problem was not due to cues such as position, color, or form,
Robinson (1933) presented
two kinds of trials: with one stimulus odd (A+ B- B-) on half
the trials, the other stimulus
odd (B+ A- A- ) on the next. If the two kinds of trials are
intermixed irregularly,
successful solution might suggest a subject had learned to
abstract the “odd” concept.
Results of this early experiment using a single monkey indicated
that an animal might be
able to learn things other than associative responses. Although,
since there were only six
possible arrangements of stimuli (e.g., ABB, BAA, ABA, BAB, AAB,
BBA), it is
possible the animal had just memorized all possible
combinations.
Over the next decades many different animals’ oddity learning
abilities were
sampled using procedures similar to those used by Robinson
(1933) to determine if
oddity was a general learning phenomenon. Subsequent oddity
problems were given to
chimpanzees (Nissen & McCulloch, 1937), monkeys (Strong
& Hedges, 1966), rats
(Wodinsky & Bitterman, 1953), canaries (Pastore, 1954), and
cats (Boyd & Warren,
1957; Warren, 1960) all of which were able to correctly solve an
oddity learning
problem. Although, cats were able to solve the oddity problem in
some studies (Boyd &
Warren, 1957; Warren, 1960), there was also evidence for cats’
inability to learn the
oddity problem (Strong & Hedges, 1966).
The success of nonhuman animals in oddity problems spurred
interest in whether
young children, ages four to seven years of age, also could
solve an oddity problem.
Children have been presented with the oddity problem with verbal
instructions (Lipsitt &
Serunian, 1963; Gollin & Shirk, 1966) as well as with
nonverbal instructions (Overman,
Bachevalier, Miller, & Moore, 1996) in order to simulate the
oddity experiments using
-
18
nonhuman animal subjects. Experiments with child subjects
provided evidence that a
child’s proficiency on an oddity problem is correlated with age
as well as type of
instruction given. Interestingly, children presented with an
oddity problem and given only
nonverbal instructions were unable to solve the problem with
adult level proficiency
(zero incorrect responses) until the age of seven (Overman et
al, 1996).
At the same time that researchers were exploring oddity learning
abilities in
different species, they were also exploring the effects of
manipulating the number of
incorrect alternatives and placement of reward. The rate of
learning in an oddity task has
been shown to be facilitated by an increase in the number of
incorrect alternatives in
chimpanzees (Nissen & McCulloch, 1937), canaries (Pastore,
1954), and pigeons
(Zentall, Hogan, Edwards, & Hearst, 1980). Other studies
have manipulated placement of
reward to determine whether there is an effect on learning if
reward is presented across
trials on the odd stimulus in all three positions, left, middle,
and right (e.g., ABB, ABA,
AAB, BAA, BAB, BBA), or if reward is presented across trials on
only the left or the
right stimulus (ABB, BAA, BBA, AAB). The latter placement of
reward has been
regarded as the easier of the two (Moon & Harlow, 1955; Boyd
& Warren, 1957; Zentall,
Hogan, Edwards, & Hearst, 1980; Bailey & Thomas,
1996).
Rats, thus far, have succeeded in learning oddity using visual
stimuli (Wodinsky,
& Bitterman, 1953), but have failed to exhibit positive
transfer of oddity using odor
stimuli (Thomas & Noble, 1988; Bailey & Thomas, 1996).
Although, Langworthy and
Jennings (1972) claim to have successfully taught rats to learn
an oddity problem using
odor, odor is very difficult to control and might have provided
discriminative cues for the
rats. Also, failing to exhibit oddity learning after thousands
of training trials are raccoons
(Strong & Hedges, 1966). Octopuses (Boal, 1991) were unable
to perform at better than
chance levels when every trial was novel. Alternatively, pigeons
(Zentall, Hogan,
Edwards, & Hearst, 1980), gulls and ravens (Benjamini,
1983), a goat (Roitberg & Franz,
2004), and a sea lion (Hille, Dehnhardt, & Mauck, 2004),
performed at above-chance
levels. Although pigeons were able to perform well on a transfer
test with novel stimuli,
there was marked impairment in performance from the last
training session to the transfer
-
19
trials (Zentall, Hogan, Edwards, & Hearst, 1980). With
extensive training, both gulls and
ravens were able to successfully perform better than chance on
the first trial of a transfer
test (Benjamini, 1983). Only one goat of 17 was able to
demonstrate transfer on the first
trial (Roitberg & Franz, 2004). A sea lion was able to
successfully transfer to novel
stimuli, however, only after 2,600 training trials (Hille,
Dehnhardt, & Mauck, 2004).
Given the mixed results from a variety of studies on oddity
learning, an animal’s
ability to learn oddity may not be the result of a single
process, although there always is a
possibility that the experimental designs used to test for
oddity are somehow flawed.
Regardless, there is more evidence for oddity learning in
vertebrates than invertebrates.
Only one study has been conducted examining oddity learning in
an invertebrate, octopus
(Boal, 1991). More research needs to be conducted in order to
explore whether
invertebrates are capable of oddity learning.
The aim of this work is to determine whether honeybees are able
to solve an
oddity learning task using a simultaneous oddity procedure. Four
experiments are
described here. Previous pilot work on the oddity problem in
this laboratory with
honeybees using the traditional simultaneous procedure produced
mixed results (Personal
communication with P.A. Couvillon: Couvillon, unpublished). The
current work differs
from the previous work in that new stimuli were used. The new
stimuli are composed of
two-color compounds in a pinwheel arrangement, or six
alternating color wedges, on a
Petri dish. These were developed to increase the salience of the
stimuli and to better
promote attention by the bees. The bees successfully
discriminated pairs of the 2 two-
colored stimuli in a two-choice procedure, and they did so
readily whether the pairs had
one color in common or no color in common. It seemed reasonable,
then, to use the new
stimuli in oddity studies.
In Experiment 1, honeybees were trained in an oddity problem
using one odd and
two identical nonodd colored Petri dishes. Bees were rewarded
with sucrose solution for
choosing the odd stimulus. In Experiment 2, honeybees again were
trained with the
colored Petri dish stimuli and were given an oddity problem
using a trial-unique method,
different stimuli on each trial. In Experiment 3, honeybees also
were trained on an oddity
-
20
problem using the trial-unique method of Experiment 2. However,
the stimuli of
Experiment 3 were displayed on a computer monitor. In Experiment
4, the stimuli were
also presented on a computer monitor, but the bees were required
to choose the nonodd
stimulus.
-
21
Experiment 1
Introduction
The aim of Experiment 1 was to determine whether honeybees are
able to learn a
simultaneous oddity problem using Petri dishes with colored
surfaces. The plan was to
first try to demonstrate oddity learning and then to analyze it
in subsequent experiments.
In each trial, bees were presented with two nonodd stimuli and
one odd stimulus. In order
to receive a sucrose reward, the bees had to choose the stimulus
that was different from
the other two. Alternatively, stevia, which is an aversive
substance, was placed on the
two nonodd stimuli in each trial. This experiment included two
groups of bees; one group
was trained with a pair of two-color compounds with a common
color, and the other
group with a pair of two-color compounds with no common color.
It seemed reasonable
that the discrimination of oddity might be easier with no color
in common.
Subjects
The subjects were 24 honeybees (Apis mellifera) never used in
prior experiments.
They were captured at feeders containing 10-20% sucrose solution
near the hives in back
of the Békésy Laboratory at the University of Hawai‘i at Mānoa.
The subjects were
assigned to two groups, No Color in Common group (NCC) and One
Color in Common
group (OCC). The bees in each group were trained individually in
a single daily session
lasting from one to several hours.
-
22
Apparatus & Stimuli
Bee enters from outside
Experimenter observes from inside lab
Figure 1. Apparatus used for Experiments 1 and 2.
The main apparatus used for training, shown in Figure 1, was a
wooden enclosure,
61 cm wide, 61 cm high, and 61 cm deep, recessed in the exterior
wall of a laboratory
window. Two sliding Plexiglas partitions separated the interior
of the laboratory from the
exterior, which prevented both unwanted entrance of the bees
into the room and
permitted observation by the experimenter.
The stimuli used were Petri dishes 5.5 cm in diameter. Patterns
of six equally
segmented triangles made of vinyl plastic with a matte-finish
were placed on each dish
using silicon sealant. Dishes displayed two of the following
colors in alternating
sequence: blue (B), green (G), orange (O), and yellow (Y). All
four colors had been
shown in previous experiments conducted in this laboratory to be
highly discriminable
and equally preferred. In addition, previous color
discrimination experiments conducted
in the laboratory have shown that each combination of two-color
compound stimuli can
be discriminated. There are six two-color compound stimuli and
fifteen total pairs of two-
5.1 cm 5.1 cm
-
23
colored compounds. Due to the nature of the fifteen combinations
of two-color
compound stimuli, three combinations have no color in common and
twelve have one
color in common. For this reason, two groups of bees were
assessed, group No Color in
Common (NCC) and group One Color in Common (OCC). In each trial,
three stimuli,
two identical and one different, were presented simultaneously,
positioned in the center
of the floor of the wooden apparatus, 5.1 cm apart edge-to-edge
parallel to the window
edge, as shown in Figure 1.
Procedure
Pretraining: A forager bee that was collected in a matchbox from
the feeder was
brought into the laboratory and released on a pretraining
stimulus (one of the two stimuli
to be used in training) in the middle of the floor of the wooden
enclosure. A 100-μl drop
of 50% sucrose was placed on top of the pretraining stimulus. As
the bee was feeding it
was marked on the thorax with colored enamel for identification
purposes. The honeybee
drank until replete and then flew to the hive (usually returning
in 3 to 5 minutes). A timer
was used to record the roundtrip return time of the experimental
bee. If the honeybee
returned to the window, it found another drop of 50% sucrose on
the second pretraining
stimulus (the other stimulus to be used in training). If it did
not return, either it was
captured again from the feeding station, or another bee was
collected. This process was
repeated until a bee reliably returned to the pretraining
stimuli.
Training: After the pretraining phase, training began. In each
training trial, the
bee found three stimuli, two identical and one different. The
two identical, nonodd
stimuli always had a100-μl drop of 10% stevia solution placed
centrally on top of the
stimulus. Previous experiments conducted in this laboratory
provided evidence that stevia
is aversive to bees. For this reason, stevia was used in
training to “punish” bees for
choosing an incorrect stimulus and to ensure that each stimulus
had a drop placed on top
for perceptual similarity. (Pilot work in this laboratory showed
that honeybees cannot
-
24
discriminate a stevia drop from a sucrose drop on a Petri dish
without tasting.) The one
different, odd stimulus always had a 100-μl drop of 50% sucrose
solution placed on top.
Sucrose was used on the odd stimulus to reinforce the bees for
choosing the correct
stimulus.
Only a bee’s initial choice in each trial was recorded and used
for analysis.
Correct responses were those where the bee landed and proceeded
to drink the sucrose on
the odd stimulus. If the bee chose correctly, it flew back to
the hive, regurgitated its
meal, and returned for another trial. A correction procedure was
used here; if the bee
chose incorrectly, it was allowed to fly (or in rare
circumstances walk) to its next choice
until the correct choice was made. If the bee touched or drank
the stevia from either of
the nonodd stimuli, an error was recorded.
Individual bees in both groups received a single session that
consisted of 15 trials,
and each session was balanced for position. Six possible spatial
arrangements of stimuli
(e.g., ABB, ABA, BAB, AAB, BBA, BAA) were used to prevent
position discrimination
and to promote oddity discrimination. Reward occurred five times
in each position (left,
middle, right) in a quasi-random sequence over the 15 training
trials. For each bee, on
seven of the trials one of the two-color compounds was odd, and
on the other eight trials
the other two-color compound was odd, with trials of the two
kinds intermixed in a quasi-
random sequence. Identical two-color compounds could be rewarded
on two trials in a
row but only in different positions.
Results & Discussion
The performance of the two groups is plotted in Figure 2 in
terms of the
proportion of bees in each group that chose correctly across
training trials; the line at .33
indicates chance. The trial-by-trial results for group NCC are
shown in the left panel and
for group OCC in the right panel. Both groups show better than
chance performance. The
mean proportion correct choices over 15 trials for group NCC was
.42 and was
significantly greater than chance [t(11) = 2.28, p < .05].
The mean proportion correct
-
25
choices over 15 trials for group OCC was .50 and was
significantly greater than chance
[t(11) = 3.90, p < .01].
Figure 2. Results for the two groups trained in Experiment
1.
Although performance in the OCC group appears to be better than
that of the
NCC group, there was no significant difference in the overall
proportion correct choices
between the two groups [t(22) = -1.40, p > .05]. These
results suggest that honeybees
treat the two-color stimuli as compounds. The results do suggest
that honeybees are
responding on the basis of oddity. However, the performance of
both groups also
suggests that the discrimination is difficult.
It is common in vertebrate studies of choice discrimination
learning in difficult
problems to further analyze the results to determine if the
subjects used systematic
strategies (Moon & Harlow, 1955). Therefore it seemed
prudent to further analyze the
bees’ choices to determine first if there was a significant
position or color preference. It
-
26
also seemed prudent to look for any tendency on a given trial to
choose the rewarded
position or the rewarded stimulus of the preceding trial.
Position preference, the tendency to choose one position (left,
middle, right) more
than another position, was analyzed. If bees do not have a
position preference, it is
expected that they will have equal frequencies of initial
position choices across all
training trials. However, a chi-square test for equal
frequencies showed significant
preference in both groups [NCC, χ 2 (2) = 7.59, p < .05; OCC,
χ 2 (2) = 12.69, p < .005].
In group NCC, there was a tendency to choose the middle
position, and, in group OCC, a
tendency to choose the left position.
Stimulus preference, that is, the tendency to prefer one color
compound over the
other, was also analyzed. It is expected that bees with no
stimulus preference will choose
each two-color compound 50% of the time. Analysis of the bees’
stimulus preference
using a one-sample t-test yielded mixed results. Overall, group
NCC did not have a
stimulus preference [t(11) = .522, p >.05], although 5 out of
12 bees clearly demonstrated
a preference indicated by 5 more choices to one stimulus over
another. On the other hand,
group OCC overall had a significant stimulus preference [t(11) =
2.45, p < .05], although
only 2 out of 12 bees had a clear preference. These results,
taken together, suggest that
some of the individual bees in each group had a stimulus
preference, that is, a nonrandom
choice of stimulus on any given trial.
Next, the tendency of position reward following was analyzed.
Position reward
following is when a subject chooses the same position on trial
n+1 that had been
rewarded on trial n. For example, if on trial n the left
position was rewarded, position
reward following on trial n+1 would be indicated by choice of
the left position. To
determine whether either group of bees had a tendency across
trials for position reward
following, the bees’ proportion of initial choices following a
previously rewarded
position was analyzed; chance should be .33. The mean proportion
for NCC was .30 and
was not significantly different than chance [t(11) = -1.05, p
> .05]. The mean proportion
for group OCC was .28 and was not significantly different than
chance [t(11) = -1.34, p >
.05]. Neither group chose a position based on immediate prior
reward of that position.
-
27
Finally, stimulus reward following was analyzed. Stimulus reward
following was
analyzed exactly the same way as position reward following. For
example, if on trial n
stimulus A was rewarded, stimulus reward following on trial n+1
would be indicated by
choice of stimulus A. To assess the possibility that the
honeybees could be making
responses based on stimulus reward following, the bees’
proportion of initial choices of
the previously rewarded stimulus was analyzed; chance should be
.50. Analysis of both
groups did not yield significant results. The mean proportion
for group NCC was .56 and
was not significantly different than chance [t(11) = 1.61, p
> .05]. The mean proportion
for group OCC was .43 and was not significantly different than
chance [t(11) = -2.05, p >
.05]. Neither group chose a stimulus based on immediate prior
reward of that stimulus.
In summary, both groups show a small, but significant tendency
to choose the odd
stimulus across the 15 training trials. Position preferences
were found in both groups.
Stimulus preferences, however, were only shown in group OCC and
not in group NCC.
Analysis of position and stimulus reward following did not yield
any significant results.
The small tendency to choose the odd stimulus could be due to
conditional
discrimination learning or it could be due to oddity learning.
Conditional discrimination
learning needs to be ruled out if the results are to be
interpreted as evidence for oddity
learning. The oddity problem presented in this experiment can be
learned based on
conditional discrimination; if BB choose A and if AA choose B.
Traditionally, as in
matching-to-sample (MTS) procedures, a nonrewarded transfer test
is needed to
demonstrate concept learning and to rule out conditional
discrimination learning. To rule
out conditional discrimination learning in an oddity problem, a
nonrewarded transfer test
to novel stimuli can be used. For example, if the bees were
trained on A+ B- B- (reward
indicated by +) and B+ A- A-, then transfer could be to C+ D-
D-. On the transfer test,
choice of stimulus C would indicate successful transfer. A
transfer test may not be
feasible here because the performance in training was so
variable. Another commonly
recognized method for testing for concept learning is a
trial-unique procedure (Wright,
Cook, Rivera, Sands, & Delius, 1988), where every trial is
different. It should be noted
that there is almost no evidence in the vertebrate literature
for successful trial-unique
-
28
performance in any discrimination including oddity and MTS.
However, work on MTS
and NMTS with bees (Shishimi, doctoral dissertation, in
progress) found successful MTS
and NMTS with a trial-unique procedure. On that basis, it seemed
reasonable to try an
oddity problem with a ‘trial-unique’ procedure, where every
trial is a transfer test, rather
than attempt a replication of Experiment 1 followed by a
transfer test with novel stimuli.
-
29
Experiment 2
Introduction
The results of Experiment 1 suggest a weak oddity learning
effect, however, it is
difficult to interpret. The aim of this experiment was to try a
trial-unique procedure, an
alternative to a transfer test, using the same kind of stimuli
as used in Experiment 1.
Every trial consists of a different combination of stimuli
(e.g., ABB, DCC, EAA, CFF,
BEE, FDD, EBB, DBB, FAA, EDD, CBB, FEE, ADD, BFF, CAA) such that
in each
trial, bees were presented with two nonodd stimuli and one odd
stimulus. Better than
chance choice performance would provide strong evidence of
oddity learning.
Subjects
The subjects were 12 honeybees (Apis mellifera) never used in
prior experiments.
They were captured at feeders containing 10-20% sucrose solution
near the hives in back
of Békésy Laboratory at the University of Hawai‘i at Mānoa. Each
subject was trained
individually in a single daily session lasting from one to
several hours.
Apparatus & Stimuli
The same apparatus and stimuli used for training in Experiment 1
(see Figure 1)
were used here. Two two-color compound arrangements of four
possible colors produce
15 unique combinations allowing for each trial in a 15-trial
session to be unique. As
Wright, Cook, Rivera, Sands, & Delius (1988) note, a
trial-unique procedure may
enhance the speed of learning acquisition by providing numerous
training exemplars.
All bees received a single session that consisted of 15 training
trials. There were
four different trial sequences of the stimulus configurations,
each used for three subjects.
The sequences were constructed so that successive trials did not
share any identical
-
30
stimuli, ensuring that the same two-color compound was not
rewarded twice in a row (see
Table 1 for a sample trial sequence). Reinforced two-color
compounds were never
presented in the same position more than once in the 15-trials.
In each trial, three stimuli
-- two identical and one different -- were presented
simultaneously, positioned in the
center of the floor of the wooden apparatus, 2.55 cm apart
edge-to-edge parallel to the
window edge. (Note that to reduce the likelihood of position
biases in the bees’ choices,
the distance between the stimuli was decreased from the 5.1 cm
used in Experiment 1.)
Procedure
Pretraining: Pretraining was exactly the same as in Experiment
1, except the
pretraining stimulus consisted of four equal segments of each
color to be used in training
(blue, yellow, orange, and green).
Training: Training was the same as in Experiment 1, except that
different pairs of
two-color stimulus compounds were presented on each training
trial (see Table 1).
Results & Discussion
The performance of the bees is plotted in Figure 3 in terms of
the proportion of
bees that chose correctly across training trials; the line at
.33 indicates chance. The mean
proportion correct choices over 15 trials was .49, which is
significantly greater than
chance [t(11) = 4.55, p = .001]. Taken at face value, this
result is evidence for oddity
learning in honeybees. Nonetheless, given the position
preference of the bees in
Experiment 1, it was necessary to further analyze the bees’
choices to determine if there
was a significant position preference or a tendency for position
reward following.
-
31
Figure 3. Results of Experiment 2.
Position preference, the tendency to choose one position (left,
middle, right) more
than another position, was analyzed. If bees do not have a
position preference, it is
expected that they will have equal frequencies of initial
position choices across all
training trials. A chi-square test for equal frequencies showed
no significant position
preference [χ 2 (2) = 2.50, p >.05], suggesting that the bees
have no systematic tendency
to choose any one of the three positions.
Position reward following also was analyzed to determine whether
the bees had a
tendency on any given trial to follow the position rewarded on
the preceding trial; chance
was set at .33. The mean proportion of position reward following
was .25 which was
significantly less than chance [t(11) = -3.07, p
-
32
not clear if the bees’ position-switching is due to active
avoidance of the previously
rewarded position. Since the trial sequences were constructed
(see Table 1) such that the
position of the rewarded stimulus was rarely the same on two
successive trials, the bees
may simply have learned not to revisit the rewarded position of
the previous trial.
In conclusion, the results of this experiment provide strong
evidence of oddity
learning in honeybees. Analysis of position preference did not
yield significant results,
although there was a significant tendency to not choose the
previously rewarded position.
It is not possible to determine if the ‘switching’ behavior
exhibited by the bees is due to a
nonrandom choice tendency or is an artifact of learning to
choose on the basis of oddity.
The oddity learning found in this experiment cannot be explained
by associative learning.
Since every trial is unique, bees are not able to learn the
oddity problem based on
conditional discrimination, although it is possible that the
bees have a preference for odd
or “different” stimuli. Therefore, it is important also to
demonstrate that honeybees can
solve a nonoddity problem, that is, reward for choice of one of
the two nonodd stimuli.
In the subsequent experiments reported here, a computer monitor
with
PowerPoint generated images is used to display the stimuli. The
new stimulus display
method provides more options for generating stimuli and for
manipulating their
presentation. Experiment 3 is a replication of Experiment 2 with
the computer-generated
stimuli. If the replication proved to be successful, the plan
was to then conduct a
nonoddity experiment with the same computer-generated
stimuli.
-
33
Experiment 3
Introduction
The aim of this experiment was to replicate Experiment 2 using a
computer
monitor to generate stimuli, which allows for easier
manipulation and more options for
spatial presentation of stimuli, as well as for varying the
number of stimuli. This
experiment used the same trial-unique procedure that was used in
Experiment 2. In each
trial, bees were presented with two nonodd stimuli and one odd
stimulus.
Subjects
The subjects were 12 honeybees (Apis mellifera) never used in
prior experiments.
They were captured at feeders containing 10-20% sucrose solution
near the hives in back
of Békésy Laboratory at the University of Hawai‘i at Mānoa. Each
subject was trained
individually in a single daily session lasting from one to
several hours.
-
34
Apparatus & Stimuli
Bee enters from outside
Experimenter observes from inside lab
Figure 4. Apparatus used for Experiments 3 and 4.
The main apparatus used for training, shown in Figure 4, was a
wooden enclosure,
61 cm wide, 61 cm high, and 61 cm deep, recessed in the exterior
wall of a laboratory
window. A flat screen 15-inch computer monitor with a glass
surface replaced the floor
of the apparatus. The monitor was covered by a thin piece of
plywood (painted gray) with
an opening cut out to expose the display. The visible display
surface was recessed into
the plywood covering 1.5 cm. The measurements of the visible
display surface were 21.6
cm by 29.2 cm. Two sliding Plexiglas partitions separated the
interior of the laboratory
from the exterior, which both prevented unwanted entrance of the
bees into the room and
permitted observation by the experimenter.
The shape, size, and distance of the two-color compound stimuli
used in
Experiment 2 were approximated with a series of slides generated
in PowerPoint. The
-
35
displayed stimuli were 6.5 cm by 5 cm in diameter and were
positioned 2 cm apart edge-
to-edge parallel to the window edge, as shown in Figure 4.
Patterns of six equally
segmented wedges were displayed on each stimulus. Stimuli
displayed two of the
following colors in alternating sequence: Purple (P), green (G),
yellow (Y), and white
(W), all of which had been used in previous discrimination
experiments conducted in this
laboratory using computer-generated images and shown
discriminable. The colors were
created in Microsoft Paint by specifying values for yellow,
white, green, and purple. The
values are as follows: yellow (red = 255, green = 255, blue = 0;
hue = 40, saturation =
240, luminosity = 120), white (red = 255, green = 255, blue =
255; hue = 160, saturation
= 0, luminosity = 240), green (red = 0, green = 153, blue = 0;
hue = 80, saturation = 240,
luminosity = 60), purple (red = 153, green = 0, blue = 153; hue
= 200, saturation = 240,
luminosity = 72). To facilitate the bees’ detection of the
stimuli, the background of the
PowerPoint slides was black to maximize contrast with the
colored stimuli.
Procedure
Pretraining: Pretraining was the same as in Experiment 2, that
is, exposure to a
stimulus compound of the four colors to be used in training. The
pretraining stimulus was
displayed as equal segments of yellow, white, green, and
purple.
Training: Training was the same as in Experiment 2, except that
a 15% salt
solution was used on the nonrewarded stimuli instead of the
stevia solution used before.
Prior experiments in this laboratory indicated that a drop of
stevia can be discriminated
from a drop of sucrose on a computer monitor, but a drop of salt
solution cannot be
discriminated from a drop of sucrose. Also different from
Experiments 1 and 2 were the
recording measurements used to indicate correct and incorrect
choice. Since the bees
often walked on the computer monitor, a correct choice was
measured as drinking from
the odd stimulus. An incorrect choice was measured as drinking
from either of the
nonodd stimuli. (See Table 1 for a sample trial-unique
sequence.)
-
36
Results & Discussion
The performance of the bees is plotted in Figure 5 in terms of
the proportion of
bees that chose correctly across training trials; the line at
.33 indicates chance. The mean
proportion correct choices over 15 trials was .51, which is
significantly greater than
chance [t(11) = 5.05, p < .000]. This result replicates that
of Experiment 2 and provides
additional evidence for oddity learning with trial-unique
stimuli. As in Experiment 2, the
bees’ choices were further analyzed to determine if there was a
significant position
preference or a tendency for position reward following.
Figure 5. Results for Experiment 3.
-
37
Position preference, the tendency to choose one position (left,
middle, right) more
than another position, was analyzed. If bees do not have a
position preference, it is
expected that they will have equal frequencies of initial
position choices across all
training trials. A chi-square test for equal frequencies showed
a significant position
preference [χ 2 (2) = 9.72, p < .01], suggesting that the
bees have a systematic tendency to
choose left position.
Position reward following also was analyzed to determine whether
the bees had a
tendency on any given trial to follow the position rewarded on
the preceding trial; chance
was set at .33. The mean proportion of position reward following
was .31 which was not
significantly different than chance [t(11) = -.519, p > .05].
The bees did not chose a
position based on immediate prior reward of that position.
In conclusion, the results of this experiment again provide
evidence of oddity
learning in honeybees, here with computer-generated stimuli. In
the next experiment
honeybees were trained in a nonoddity problem, where choice of
either of the two
identical stimuli was rewarded and choice of the odd stimulus
was not rewarded. If the
oddity learning seen in the previous experiments was enhanced by
an unlearned
preference for oddity, then honeybees should not be able to
learn the nonoddity problem,
or, at least, learning should appear to be more difficult than
in the oddity problem.
-
38
Experiment 4
Introduction
The aim of this experiment was to determine whether honeybees
could learn to
choose on the basis of nonoddity. This experiment used the same
trial-unique procedure
as used in Experiment 3 except a reward was provided on both of
the nonodd stimuli and
salt solution was placed on the odd stimulus. The results of
this experiment are important
for an understanding of the oddity learning demonstrated in
Experiments 2 and 3. If
oddity learning is based mostly on an unlearned preference for
oddity in honeybees, then
the nonoddity problem should be difficult, if not impossible.
However, if the bees also
solve the nonoddity problem, the results for both oddity and
nonoddity may be based on
concept learning, which, as noted above, refers to learning that
is not readily accountable
for by associative learning principles.
Subjects
The subjects were 12 honeybees (Apis mellifera) never used in
prior experiments.
They were captured at feeders containing 10-20 % sucrose
solution near the hives in back
of Békésy Laboratory at the University of Hawai‘i at Mānoa. Each
subject was trained
individually in a single daily session lasting from one to
several hours.
Apparatus & Stimuli
The same apparatus used for training in Experiment 3 was used
here, and the
stimuli also were the same.
-
39
Procedure
Pretraining: Pretraining was the same as in Experiment 3.
Training: Training was the same as the trial-unique training of
Experiment 3,
except 50% sucrose reward was provided on the two nonodd stimuli
and an aversive 15%
salt solution on the odd stimulus. (See Table 1 for a sample
trial-unique sequence.)
Results & Discussion
The performance of the bees is plotted in Figure 6 in terms of
the proportion of
bees that chose correctly across training trials; the line at
.66 indicates chance. Chance is
now .66 because the bees are given two rewarded stimulus
compounds. The mean
proportion correct choices over 15 trials was .75, which is
significantly greater than
chance [t(11) = 2.88, p < .05]. This result clearly indicates
that honeybees can learn a
nonoddity problem. As in the previous experiments, the bees’
choices were further
analyzed to determine if there was a significant position
preference or a tendency for
position reward following.
-
40
Figure 6. Results for Experiment 4.
Position preference, the tendency to choose one position (left,
middle, right) more
than another position, was analyzed. If bees do not have a
position preference, it is
expected that they will have equal frequencies of initial
position choices across all
training trials. A chi-square test for equal frequencies did not
show a significant position
preference [χ 2 (2) = 1.58, p > .05], suggesting that the
bees do not have a systematic
tendency to choose any one of the three positions.
Position reward following also was analyzed to determine whether
the bees had a
tendency on any given trial to follow the position rewarded on
the preceding trial; chance
was set at .33. The mean proportion of position reward following
was .31 which was not
significantly different than chance [t(11) = -.56, p > .05].
The bees did not chose a
position based on immediate prior reward of that position.
-
41
In conclusion, the results of this experiment provide evidence
of nonoddity
learning in honeybees. It is worth noting that the likelihood of
choosing correctly by
chance is very high, .66. Even in simple choice discrimination
problems honeybees rarely
choose correctly 100% of the time, often achieving about .90
proportion of correct
choice. It is noteworthy that the bees in this experiment
performed significantly better
than chance (.75) given the possibility of a ceiling effect. The
fact that the bees readily
solved the nonoddity problem argues against the possibility that
the bees have an
unlearned preference for oddity. That honeybees show both oddity
and nonoddity
learning with trial-unique stimuli is a compelling case for
concept learning.
-
42
General Discussion
The purpose of the four experiments reported here was to
determine if honeybees
could learn an oddity problem. Robinson (1933) first established
that a simultaneous
oddity problem was one way to study concept learning in nonhuman
animals. These
experiments are the first serious attempt to examine whether
honeybees can learn oddity.
Experiment 1 was a variant of Robinson’s (1933) classical
2-stimulus oddity
problem, where reward occurs on stimulus A on one kind of trial
(A + B- B-) and on
stimulus B on the other kind of trial (B+ A- A-). Honeybees
given this problem were
trained with colored Petri dishes and rewarded with sucrose for
choice of the odd color
on each trial. The results showed that honeybees were able to
perform at levels greater
than chance. However, their small tendency to choose the odd
stimulus could have been
due to conditional discrimination learning rather than to oddity
learning.
Experiment 2 was conducted in order to rule out conditional
discrimination
learning. Usually to rule out conditional discrimination,
transfer trials with new stimuli
(e.g., C+ D- D- and D+ C- C-) are given to experimental subjects
after training with A+
B- B- and B+ A- A- . Correct choice on the first transfer trial
is typically taken as
evidence for oddity or concept learning. Another method is to
use unique stimuli on
every trial. The trial-unique method is regarded, in the
vertebrate literature, as the “gold
standard” to test for concept learning. Unpublished work in this
laboratory (Shishimi,
doctoral dissertation, in progress) found successful
trial-unique matching-to-sample and
non-matching-to-sample performance in honeybees. Therefore, it
seemed reasonable to
use this approach with an oddity problem. The trial-unique
method was used in
Experiment 2, and honeybees were able to perform at levels
greater than chance. These
results cannot be explained as conditional discrimination
learning and provide strong
evidence of oddity learning in honeybees.
Experiment 3 was a replication of Experiment 2 using the
trial-unique stimulus
method with new stimuli displayed on a computer screen. The
honeybees again were able
to successfully learn the oddity problem at levels greater than
chance. These results,
-
43
which replicate those of Experiment 2, again cannot be explained
as simple conditional
discrimination because the stimuli change on every trial. The
only other possible
explanation of the honeybees’ greater than chance performance is
that honeybees could
have an unlearned preference for odd or “different” items. The
issue of whether oddity
learning could be explained by such a preference still needed to
be addressed.
Experiment 4 was designed to determine if honeybees could solve
a nonoddity
problem. If the oddity learning found in Experiments 2 and 3 was
based only on a
preference for choosing the odd stimulus, then a nonoddity
problem should be difficult
for honeybees. The same procedure and trial configurations of
Experiment 3 were used in
Experiment 4, but reward occurred on the nonodd stimuli instead
of the odd stimulus.
The greater than chance performance indicates that honeybees are
able to learn a
nonoddity problem.
In summary, the experiments reported here provide compelling
evidence of both
oddity and nonoddity learning in honeybees. These results, taken
together, cannot be
explained by associative learning (conditional discrimination)
or unlearned preference for
novelty. The results do suggest that honeybees may have learned
to discriminate oddity
and nonoddity on the basis of concepts.
Traditionally, learning research conducted with invertebrates
has been primarily
concerned with exploring basic learning phenomena using analogs
of vertebrate learning
experiments. This approach was used here, that is, to explore
how honeybees perform in
oddity problems that are analogous to those used with
vertebrates. The honeybees’
successful performance here is noteworthy because vertebrate
performance in oddity
problems is highly variable, sometimes successful and sometimes
not (for example see,
Boyd & Warren, 1957; Warren, 1960; Strong & Hedges,
1966). Comparison of
experimental details between the honeybee experiments and those
of the vertebrate
experiments point to two methodological differences that may
contribute to the
differences in performance seen across species on oddity
problems.
One methodological difference is that vertebrate oddity
experiments typically
have massed trials, whereas, the above experiments with
honeybees have spaced trials.
-
44
Oddity training with pigeons, for example, is conducted over
many daily sessions with 20
to 30 trials per session. The intertrial interval typically
ranges from 10 to 30 seconds. The
oddity training with honeybees was conducted in a single session
with 15 trials. The
intertrial interval averaged 3 to 5 minutes. (Note that the
intertrial interval for honeybees
was determined by the time it took for the bee to leave the
window, deposit the sucrose at
the hive, and return to the window for another trial.) Highly
massed trials can lead to both
proactive and retroactive interference that may contribute to
the variable performance
seen across the different vertebrate species. Spaced trials may
minimize proactive and
retroactive interference, and, therefore, contribute to
honeybees’ successful performance
on the oddity and nonoddity problems. Further research needs to
be conducted to
determine the extent to which massed and spaced trials affect
performance in oddity
problems.
Another methodological difference between the vertebrate and
honeybee
experiments is the use of punishment. Typically in vertebrate
experiments, subjects are
given reward for choosing a correct stimulus and no reward for
choosing an incorrect
stimulus. In the honeybee experiments, subjects were given
reward (sucrose) for
choosing a correct stimulus but also were punished with stevia
or salt for choosing an
incorrect stimulus. The use of punishment may facilitate
learning because it may increase
the cost of incorrect choice which also could increase attention
to the stimuli.
Since the goal of comparative research with honeybees is to
compare their
performance to that of vertebrates in similar problems, it is
useful to review the
theoretical interpretations that have been discussed in the
vertebrate literature. Concept
learning has been mentioned here as a possible basis for the
successful performance of
the oddity and nonoddity problems by honeybees. However,
“concept” is only one term
that is used in the vertebrate literature, but is by no means
standard. Other terms in the
literature include, relational discrimination, stimulus
relations, relational concepts, events
and their interrelations, abstract concepts, learning relations
about relations, higher-order
relations, and generalized concepts (Cook and Wasserman, 2012).
The conflicting
definitions and terminology for “concept” may not just be a
difference of style or
-
45
phrasing. For example, both Cook and Wasserman (2012) have
conducted experiments
on concept learning in pigeons. Both experimenters use
Same/Different (S/D)
discriminations in their respective laboratories. However, Cook
(2012) uses a multi-
stimulus discrimination array to test for concepts, whereas,
Wasserman (2012) uses an
icon-based discrimination array to test for concepts. Both Cook
and Wasserman (2012)
use the same term “concept” to explain the successful
performance on the S/D
discrimination problem by their pigeons. However, Wasserman
(2012) considers another
possibility, entropy, “an information theoretic concept,”
defined operationally as “item
variability.” Wasserman (2012) has empirical evidence that an
increase in the number of
icons in his display (increase in entropy) improves performance
in S/D problems. Cook
(2012), however, fails to find such improvement, and, in fact,
finds deterioration in
performance with an increase in the number of stimuli in his
arrays. Cook and
Wasserman (2012) conclude, “…different tasks rely on different
perceptual and
conceptual processes.” Whether honeybees are using perceptual or
conceptual processes
still needs further examination. Given the similarities in
display types between the stimuli
used here and Cook’s (2012) stimuli, it is possible that
honeybees may solve the oddity
and nonoddity problem using concepts.
-
46
Table 1. Sample trial sequence for Experiment 2.
L (left), M (middle), R (right) refer to stimulus position in
the experimental window.
-
47
References
.
Avarguès-Weber, A., Dyer, A.G., & Giurfa, M. (2011).
Conceptualization of above and
below relationships by an insect. Proceedings of the Royal
Society B: Biological
Sciences, 278, 898-905.
Avarguès-Weber, A., Dyer, A.G., Combe, M., & Giurfa, M.
(2012). Simultaneous
mastering of two abstract concepts by the miniature brain of
bees. Proceedings of
the National Academy of Science, 109, 7481-7486.
Bailey, A.M., & Thomas, R.K. (1996). An investigation of
oddity concept learning by
rats. Psychological Record, 48, 447-456.
Batson, J.D., Hoban, J.S., & Bitterman, M.E. (1992).
Simultaneous conditioning in
honeybees (Apis mellifera). Journal of Comparative Psychology,
106, 114-119.
Beach, F.A. (1950). The snark was a boojum. American
Psychologist, 5, 115-124.
Benjamini, L. (1983). Studies in the learning abilities of
brown-necked ravens and
herring gulls: I. Oddity learning. Behaviour, 84, 173-194.
Bitterman, M.E. (1975). The comparative analysis of learning.
Science, 188, 699-709.
Bitterman, M.E., Menzel, R., Fietz, A., & Schäfer, S.
(1983). Classical conditioning of
proboscis extension in honeybees (Apis mellifera). Journal of
Comparative
Psychology, 97, 107-119.
Bitterman, M.E. (1996). Comparative analysis of learning in
honeybees. Animal Learning
& Behavior, 24, 123-141.
Bitterman, M.E. (2006). Classical conditioning since Pavlov.
Review of General
Psychology, 10, 365-376.
Blough, D.S. (1959). Delayed matching in the pigeon. Journal of
the Experimental
Analysis of Behavior, 2, 151-160.
-
48
Boal, J. (1991). Complex learning in Octopus bimaculoides.
American Malacological
Bulletin, 9, 75-80.
Boal, J.G. (1996). A review of simultaneous visual
discrimination as a method of training
octopuses. Biological Reviews of the Cambridge Philosophical
Society, 71, 157-
190.
Boyd, B.O., & Warren, J.M. (1957). Solution of oddity
problems by cats. Journal of
Comparative and Physiological Psychology, 50, 258-260.
Chittka, L., & Jenson, K. (2011). Animal cognition: Concepts
from apes to bees.
Current Biology, 21, R116-R119.
Cook, R.G, & Wasserman, E.A. (2012). Relational
discrimination learning in pigeons. In
T.R. Zentall & E.A. Wasserman (Eds.), The handbook of
comparative
cognition (pp. 533 -551). New York, NY: Oxford University
Press.
Corning, W.C., Dyal, J.A., & Willows, A.O.D. (1973).
Invertebrate learning: Volume 2
arthropods and gastropods. New York, NY: Plenum Press.
Corning, W.C., Dyal, J.A., & Willows, A.O.D. (1975).
Invertebrate learning: Volume 3
cephalopods and echinoderms. New York, NY: Plenum Press.
Couvillon, P.A., & Bitterman, M.E. (1985). Analysis of
choice in honeybees. Animal
Learning & Behavior, 13, 246-252.
Couvillon, P.A., & Bitterman, M.E. (1988).
Compound-component and conditional
discrimination of colors and odors by honeybees: Further tests
of a continuity
model. Learning & Behavior, 16, 67-74.
Couvillon, P.A., Lee, Y., Bitterman, M.E. (1991). Learning in
honeybees as a function of
amount of reward: Rejection of the equal-asymptote assumption.
Animal Learning
& Behavior, 19, 381- 387.
-
49
Couvillon, P.A., Hsiung, R., Cooke, A.M., & Bitterman, M.E.
(2005). The role of context
in inhibitory conditioning of honeybees. The Quarterly Journal
of Experimental
Psychology Section B: Comparative and Physiological Psychology,
58, 59-67.
Darwin, C. (1859). On the origin of species. London, UK: John
Murray. (Reprinted by
Harvard University Press, 1964).
Darwin, C. (1871). The descent of man. New York, NY: Random
House.
Descartes, R. (1637). Discourse on methods and meditations.
Translated by L.J. Lafleur.
(Reprinted by Liberal Arts Press, 1960).
Descartes, R. (1646). Letter to the Marquis of Newcastle, quoted
by Henry A.P. Torrey,
The philosophy of Descartes (New York, Henry Holt and Company,
1892), p.
283.
Giurfa, M., Eichmann, B., & Menzel, R. (1996). Symmetry
perception in an insect.
Nature, 382, 458-461.
Giurfa, M., Zhang, S., Jenett, A., Menzel, R., & Srinivasan,
M.V. (2001). The concepts of
‘sameness’ and ‘difference’ in an insect. Nature, 410,
930-933.
Gollin, E.S., & Shirk, E.J. (1966). A developmental study of
oddity-problem learning in
young children. Child Development, 37, 213-217.
Gould, J.L., Kirschvink, J.L., & Deffeyes, K.S. (1978). Bees
have magnetic remanence.
Science, 201, 1026-1028.
Hille, P., Dehnhardt, G., & Mauck, B. (2006). An analysis of
visual oddity concept
learning in a California sea lion. Learning & Behavior, 2,
144-153.
Kandel, E.R., & Hawkins, R.D. (1992). The biological basis
of learning and
individuality. Scientific American, 79-86.
Kirchner, W.H. (1993). Vibrational signals in the tremble dance
of the honeybee, Apis
mellifera. Behavioral Ecology and Sociobiology, 33, 169-172.
-
50
Kohler, W. (1918). Nachweis einfacher structurfunktionen beim
schimpansen und beim
haushuhn. Translated and condensed as “Simple structural
functions in
chimpanzees and chickens.” In W.D. Ellis (Ed.) (1969). A source
book of Gestalt
psychology. London, UK: Routledge and Kegan Paul.
Langworthy, R.A., & Jennings, J.A. (1972). Odd ball,
abstract, olfactory learning in
laboratory rats. Psychological Record, 22, 487-490.
Lipsitt, L.P., & Serunian, S.A. (1963). Oddity-problem
learning in young children. Child
Development, 34, 201-206.
Moon, L.E., & Harlow, H.F. (1955). Analysis of oddity
learning by rhesus monkeys.
Journal of Comparative and Physiological Psychology, 48,
188-194.
Morgan, C.L. (1894). An introduction to comparative psychology.
London, UK: Scott.
Nieh, J.C., & Tautz, J. (2000). Behavior-locked signal
analysis reveals weak 200-300 Hz
comb vibrations during the honeybee waggle dance. The Journal of
Experimental
Biology, 203, 1573-1579.
Nissen, H.W., & McCulloch, T.L. (1937). Equated and
non-equated stimulus situations in
discrimination learning by chimpanzees: I. Comparison with
unlimited response.
Journal of Comparative and Physiological Psychology, 23,
165-189.
Olton, D.S., & Samuelson, R.J. (1976). Remembrance of places
passed: Spatial memory
in rats. Journal of Experimental Psychology: Animal Behavior
Processes, 2, 97-
116.
Overman, W., Bachevalier, J., Miller, M., & Moore, K.
(1996). Children’s performance
on “animal tests” of oddity: Implications for cognitive
processes required for tests
of oddity and delayed nonmatch to sample. Journal of
Experimental Child
Psychology, 62, 223-242.
Pastore, N. (1954). Discrimination learning in the canary.
Journal of Comparative and
Physiological Psychology, 47, 389-390.
-
51
Pavlov, I.P. (1927). Conditioned reflexes. New York, NY: Dover.
(Reprinted in 1960).
Robertson, H.M., & Wanner, K.W. (2006). The chemoreceptor
superfamily in the honey
bee, Apis mellifera: Expansion of the odorant, but not
gustatory, receptor family.
Genome Research, 16, 1395-1403.
Robinson, E.W. (1933). A preliminary experiment on abstraction
in a monkey. Journal of
Comparative Psychology, 16, 231-236.
Rohrseitz, K., & Tautz, J. (1999). Honeybee dance
communication: Waggle run direction
coded in antennal contacts. Journal of Comparative Physiology A,
184, 463-470.
Roitberg, E., & Franz, H. (2004). Oddity learning by African
dwarf goats. Animal
Cognition, 7, 61-67.
Romanes, G. (1892). Animal intelligence. London, UK: Kegan Paul,
Trench, Trubner Co.
Scheiner, R., Page, R.E. Jr., & Erber, J. (2001). The
effects of genotype, foraging role,
and sucrose responsiveness on the tactile learning performance
of honey bees
(Apis mellifera L.). Neurobiology of learning and memory, 76,
138-150.
Shishimi, G. (In progress). Trial-unique matching and
nonmatching-to-sample in
honeybees. (Unpublished doctoral dissertation). University of
Hawai‘i at Mānoa.
Skinner, B.F. (1938). The behavior of organisms. New York, NY:
Century.
Spence, K.W. (1936). The nature of discrimination learning in
animals. Psychological
Review, 43, 427-449.
Strong, P.N., & Hedges M. (1966). Comparative studies in
simple oddity learning: I.
Cats, raccoons, monkeys, and chimpanzees. Psychonomic Science,
5, 13-14.
Tautz, J., & Rohrseitz, K. (1998). What attracts honeybees
to a waggle dance? Journal of
Comparative Physiology, 183, 661-667.
Thomas, R.K., & Noble L.M. (1988). Visual and olfactory
oddity learning in rats: What
evidence is necessary to show conceptual behavior? Animal
Learning &
Behavior, 16, 157-163.
-
52
Thorndike, E.L. (1898). Animal intelligence: An experimental
study of the associative
processes in animals. Psychological Review Monograph Supplement,
2(4), 1-8.
American Psychologist (Reprinted 1998). Vol. 53, n 10, p.
1125-1127.
Thorndike, E.L. (1911). Animal intelligence: Experimental
studies. Darien, CT:
Hafner. (Reprinted in 1970).
Tolman, E.C. (1948). Cognitive maps in rats and men.
Psychological Review, 55, 189-
208.
von Frisch, K. (1950). Bees: Their vision, chemical sense, and
language. New York, NY:
Cornell University Press.
Walker, M.M., & Bitterman, M.E. (1985). Conditioned
responding to magnetic fields by
honeybees. Journal of Comparative Physiology A, 157, 67-71.
Warren, J.M. (1960). Oddity learning set in a cat. Journal of
Comparative and
Physiological Psychology, 5, 433-434.
Wodinsky, J., & Bitterman, M.E. (1953). The solution of
oddity-problems by the rat. The
American Journal of Psychology, 66, 137-140.
Wright, A.A., Cook, R.G., & Rivera, J.J., Sands, S.F., &
Delius, J.D. (1988). Concept
learning by pigeons: Matching-to-sample with trial-unique video
picture stimuli.
Animal Learning & Behavior, 16, 436-444.
Wright, A.A., & Katz, J.S. (2006). Mechanisms of
same/different concept learning in
primates and avians. Behavioural Processes, 72, 234-254.
Zentall, T.R., Hogan, D.E., Edwards, C.A., & Hearst, E.
(1980). Oddity learning in the
pigeon as a function of the number of incorrect alternatives.
Journal of
Experimental Psychology: Animal Behavior Processes, 6,
278-299.
-
53
PDF Cover pages.11.2.12.pdfPDF Introduction.11.2.12PDF Exp.
1-4.11.2.12